For decades, the agile movement defined how modern software was built. It was a revolution in speed and collaboration, born out of frustration with rigid waterfall processes and heavy specs. I had the privilege of experiencing that transformation firsthand as a principal engineer at Xtreme Labs and later at Pivotal/VMware, where pair programming, continuous delivery, and test-driven development & deployments weren’t just buzzwords. They were culture.
Now, we’re entering a new paradigm.
The era of AI-native software development is here, and it’s reshaping how we think about product velocity, team composition, and the very definition of an iteration.
The AI Agile Mindset Shift
We often talk about AI in terms of productivity, but the more important shift is in mindset. In the AI agile era, your team isn’t bottlenecked by the noise (broken telephone) that comes with defining what a piece of software could be. It’s bottlenecked by clarity of thought. If you can describe it, you can prototype it. That democratizes creation and empowers a broader set of stakeholders to contribute to product development.
We’re no longer debating feasibility in week-long planning sessions. Instead, we’re generating, testing, and learning—sometimes all within the same afternoon. This shift means the limiting factor isn’t how fast you can code, but how clearly you can articulate the outcome you want.
Supercharging the Agile Process
Where agile shortened iteration, AI shortens the distance from idea to production so teams start with something more tangible and iterate faster from there. Powered by large language models (LLMs) and generative copilots, this shift introduces a new type of workflow where natural language becomes the starting point for building software.
Product managers, designers, and even founders no longer need to begin with detailed specs or polished wireframes. Instead, they can describe their intent in plain English and instantly receive functional prototypes, scaffolding code, or design components to refine.
At Mantle, we’ve embraced this shift as a new way of thinking about how product work gets done.
Here’s how our co-founder and Head of Product for Mantle Equity, Jarrett Quan-Hin, has been putting this AI-native approach into practice.
“Dogfooding the platform by onboarding companies has surfaced a lot of ideas for improving how cap table data gets entered into Mantle. One recurring pain point was having to flip back and forth between the data in Mantle and the original spreadsheet. My first goal was simple: let someone upload a cap table summary and instantly see that information alongside what’s already in the platform, making it easy to spot any differences.”
1. Prompt Experimentation and Tuning
Jarrett began by testing whether an LLM could reliably extract cap table summary information from a broad range of spreadsheets. Once the prompt produced consistent results, he layered in structured output schemas that closely aligned to our existing API models.
2. Implementation with Copilot
Next, he moved into VS Code and asked GitHub Copilot to generate an implementation plan. Knowing the work was extensive, he prompted Copilot to assess the plan and break it down into atomic prompt sets so he could continually refresh the context window and prevent “context rot”.
3. Iterative Feedback to a Demo-Ready State
From there, it was rapid iteration on the UX. Jarrett started with the values being displayed side-by-side, then quickly shifted to highlighting the differences between the two figures. Copilot made it easy to quickly hone in on the overall UX without pulling any other resources to help move things around.
4. Improving Feature Development
Experimenting with edge cases led to new opportunities:
- If the spreadsheet included share classes or equity plans not yet in Mantle, why not add quick-add buttons?
- If we can detect and extract that info during onboarding, why not use it to improve our onboarding flows for all customers?
- Could the same logic power a cap table round modeler, letting anyone upload a cap table and quickly model a raise?
“While it’s important not to chase every thread, the ability to quickly implement and test ideas before involving any other resources has been a massive unlock. I can prove feasibility, refine UX, and if the idea has legs, hand off prototype code for our engineers to build on.”
This isn’t just about speed. It’s about challenging the nature of collaboration itself. Designers and engineers can now work together in real time, iterating in a shared creative loop. PMs can go from idea to working interface in a single session. Engineering teams can focus on architecture, scalability, and quality.
The Rise of Prompt-Led Development
Prompt engineering has become a core skill in modern software teams. It acts as the bridge between idea and implementation, allowing teams to generate starting points instead of blank screens. With tools like GitHub Copilot, ChatGPT Codex, and Claude Code, a single well-phrased prompt can produce API endpoints, data models, or working UI logic. What used to take hours of coordination and story pointing can now be sketched, revised, and tested within a single conversation loop.
Just as test-driven development gave developers confidence to refactor rapidly, AI tooling provides confidence to prototype, discard, and reimagine faster than ever before. The cost of being wrong is nominal. That creates more space for experimentation, which ultimately leads to better products.
Some Closing Thoughts
The agile revolution made the software cycle more responsive. The AI revolution is making it generative. It’s no longer about building faster within defined constraints. It’s about rethinking the constraints entirely. The tools are here, the workflows are evolving, and the most impactful teams will be those who embrace this shift early.
If you’re building software today, you’re not just managing sprints. You’re designing prompts, shaping outcomes, and unlocking creative bandwidth across every function.
Welcome to the AI Agile era.
Dwayne Forde is a co-founder and CTO at Mantle.